verifiedmedium · 0.70updated 2026-10-15

Apple M2 Max (38c, 96GB) llamafile: Qwen-2.5-14B Q4_K_M decode

25.23 tok/s avg generation (batch-1); prefill 252.47 t/s

Fields

benchmarks/localscore-m2-max-38c-96gb-qwen25-14b-q4km-decode

Metric

decode_tok_s

Value

25.23

Unit

tok/s avg generation (batch-1); prefill 252.47 t/s

Model

qwen-2-5-14b-instruct Q4_K_M

Methodology

localscore.ai accelerator/515 measured row via siliconscore CSV.

Measured by

third_party